Derogatory Comment Classification

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 04 | Apr 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3749
Derogatory Comment Classification
Sudhanshu Chaurasia1, KMR Dayaasaagar2, Jayesh Girdhar3 ,Dhanesh.S4,Sunil Shelke5
1,2,3,4 UG Student, Dept. of Computer Engineering, Pillai College of Engineering, New Panvel, India. 5 Assistant
Professor, Dept. of Information Technology, Pillai College of Engineering, New Panvel, India.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Social media platforms have become a facet for
people’s opinion and reviews, wherein people share their
opinions on a wide variety of topics. However, people exploit
this facility to take a dig at those with whom they don't find
their opinions match with. People use this facility to post
harmful, racial, gender biased, threatfulcomments. Asaresult,
social media platforms are quickly becoming indispensable.
They often struggle to facilitate conversation, effectively
forcing many communities to shut down user comments. This
motivates us to look into this problemandbuildamodelwhich
will detect and classify derogatory comments. We will collect
the dataset from Kaggle and experiment with the help of deep
learning approaches like Naïve-Bayes, SVM, LSTM and BERT.
algorithms. Thereby, we will classify the derogatory
comments. We will compare these algorithms and conclude
which algorithm is more effective.
Key Words: Language recognition, multilingual,
offensive word, NBSVM
1. INTRODUCTION
In today’s current society, there is a big problem when it
comes to online toxicity. Internet isanopendiscussingspace
for everyone to freely express their opinions. With the
massive increase in social interactions on online social
networks, there has also been an increase of hateful
activities that exploit such infrastructure. This toxicitytends
to negatively impact how a lot of people tend to engage in
conversation and deters some from engaging in online
conversation entirely.
As a result, online platforms tend to struggle to effectively
facilitate conversations, leading many communities to limit
or completely shut down user comments. Harassment and
abuse are discouraging people from sharing their ideas and
disturbing the internet environment. Platforms struggle to
effectively facilitate conversations, leading many
communities to limit or completely shut down user
comments if it’s toxic. Motivated by thisproblem,we want to
build a technology that detects and classifies the derogatory
comment. For our model, the input will be a comment. We
will use three different modelstopredictscoresfor“toxicity”
(which is our target), “severe toxicity”, “obscenity”, “identity
attack”, “insult”, “threat”. Then Support Vector Machine
(SVM) models are often used as baselines for other methods
in text categorization .In this case Naive Bayes -SVM (NVB-
SVM) provides more accuracy for further classification.
2. Literature Survey
A. Base-Line Model:Naive NVB-SVM was developed by Sida
Wang and Christopher Manning. Bayes (NB) and Support
Vector Machine (SVM) models areoften usedasbaselinesfor
other methods in text categorization .In this case Naive
Bayes -SVM (NVB-SVM) provides more accuracy for further
classification.
B. BERT: which stands for Bidirectional Encoder
Representations transformers. Bidirectional Encoder
Representations from Transformers (BERT) is a NLP model
that was designed to pretrain deep bidirectional
representations from unlabeled text and, after that, be fine-
tuned using labeled text for different NLP tasks [7]. That
way, with BERT model,wecancreatestate-of-the-artmodels
for many different NLP tasks [7]. We can see the results
obtained by BERT in different NLP tasks at [7].
C. Toxic Comment Detection using LSTM This papers
authors have developed a LSTM model to classify speech as
hate or not with an accuracy of 95% accuracy. LSTM stands
for long short-term memory and is an improvement over a
normal RNN. The model not only classifies a given sentence
as toxic or non-toxic but also gives the percentage of toxicity
or non-toxicity of the given sentence.
D. Research on Text Classification Based on CNN and
LSTM: In this paper a new model is developed by combining
a LSTM and CNN which outperforms the performance of
both the algorithms individually
2. 1 Summary of Related Work
The summary of methods used in literature is given in Table
1.
Literature MODEL Accuracy
Sida Wang, Baselines and
Bigrams: Simple, Good
Sentiment and Topic
Classification,
2020
NVB-SVM 87%
Hong Fu, Social Media Toxicity
Classification Using Deep
Learning ,2021
BERT 98%

Krishna Dubey, ToxicComment
Detection using LSTM ,2020
LSTM 95%
2019, YuandongLuanResearch
on Text Classification Based on
CNN and LSTM
LSTM-
CNN
98%
3. Proposed Architecture
Our research is aimed at derogatory terms in a text that the
user provides as input. Because social media has grown in
popularity in recent years, there are people who engage in
some type of online material (for example, articles) aimed
towards a person or a group, etc. So, using natural language
processing or machine learning methods, our program will
evaluate the text and refer back to the dataset to determine
the language of the input text as well as whether or not it
contains any derogatory terms.
3.1 System Architecture
The proposed system architecture is given in Figure 1
Fig. 1 Proposed system architecture
B. Block 1 Description: upload the data set which has been
taken from Kaggle . It containsWikipedia commentswhichis
classified into Obscene, severely toxic, threat, identity hate
and insult.
C. Block 2 Description: We will splitthedatasetintotraining
and testing sets . we will split the dataset into 60% for
training and 40% for testing and can be increased if there is
a demand for increasing the accuracy.
D. Block 3 Description: This datasetcontainsdata which has
special, characters, useless words and links etc. We will be
filtering them out .
E. Block 4 Description: Tokenization is an important step
when working with text data. We will use this process to
split the sentences into smaller units called tokens.
F. Block 5 Description: We will use Naïve Bayes-SVM model
to create a base-line model so as to improve the accuracy in
detecting and classifying the comments.
G. Block 6 and 7 Description: We will feed the output of the
base line model to each of these models separately. BERT
and LSTM will be able to classify and predict derogatory
comments with higher accuracy rate.
H.Block 8 Description: We will compare the results
generated by LSTM and BERT models in detecting
derogatory comments .
3. Requirement Analysis
The implementation detail is given in this section.
3.1 Software
We will be using streamlit for making the web app. We will
be using python for developing the classification model. We
will be required to import different libraries to make the
classification models .
3.2 Hardware
We will be requiring Intel core i-5 generationastrainingand
testing requires a high-performance processor. We will be
requiring GeForce 1650 or above.
3.3 Dataset and Parameters
The Conversation AI team, a research initiative founded by
Jigsaw and Google are working on tools to help improve
online conversation. One area of focus is the study of
negative online behaviors, like toxic comments (i.e.
comments that are rude, disrespectful or otherwise likely to
make someone leave a discussion). So far they’ve built a
range of publicly available models served through the

Perspective API, including toxicity. But the current models
still make errors, and they don’t allow users to select which
types of toxicity they’re interested in finding (e.g. some
platforms may be fine with profanity, but not with other
types of toxic content). A multi-headed model that’s capable
of detecting different types of toxicity likethreats,obscenity,
insults, and identity-based hate is developed. An example of
a data source by Jigsaw is given in Fig. 3.1.
Fig. 3.1 Kaggle Toxic dataset by jigsaw
3. CONCLUSIONS
To sum up our work, we implemented two machine learning
models, namely, LSTM and BERT. We utilized information
from data visualization to preprocess data. BERT can
produce state-of-the-art work with only one output layer.
We trained BERT with only one epoch and got results better
than the LSTM model.
For future work, we would like to try waystofixtheproblem
of imbalanced data. BERT model brought us surprisingly
good results and we would like to explore it if we have more
time.
ACKNOWLEDGMENT
It is our privilege to express our sincerest regards to our
supervisor Dr Sunil Shelke for the valuable inputs, able
guidance, encouragement, whole-hearted cooperation and
constructive criticism throughout the duration of this work.
We deeply express our sincere thanks our Head of the
Department Dr. Sharvari Govilkar and our Principal Dr.
Sandeep M. Joshi for encouraging and allowing us to
presenting this work.
REFERENCES
1.Sida Wang, Baselines and Bigrams: Simple, Good
Sentiment and Topic Classification,2020
2. Hong Fu, Social Media Toxicity Classification Using
DeepLearning,2021
3. Krishna Dubey, Toxic Comment Detection usingLSTM,
Third International Conference on Advances in
Electronics , 2020
4. Yuandong Luan Research on Text Classification Based
on CNN and LSTM,International Conference on Artificial
Intelligence and Computer Applications,2019

Derogatory Comment Classification

Recommended

Recommended

More Related Content

Similar to Derogatory Comment Classification

Similar to Derogatory Comment Classification (20)

More from IRJET Journal

More from IRJET Journal (20)

Recently uploaded

Recently uploaded (20)

Derogatory Comment Classification